&DRIFT: INTEGRATION-PARAMETERS
Sets the integration accuracy. This parameter enters in the update
of the stepsize used in drift line integration. The default value of
this parameter is chosen for chambers with a reasonably complicated
field. If the field is very simple, a smaller value should be chosen.
The value of this parameter does not bear an immediate relation
with the accuracy of the integration. Please consult the printed
version of the manual for further details.
When performing Monte Carlo drift line integration, one has the choice
between 3 methods:
- a constant time difference between 2 successive steps, selected
with the MC-TIME-INTERVAL option, where the time difference between
the steps has to be selected by the user
- a constant spatial difference between 2 successive steps, selected
with the MC-DISTANCE-INTERVAL option, where the distance in space
between the steps has to be selected by the user
- a simulation of the collision process with at each step a randomly
selected distance between 2 collisions, based on the mean free
path for the local electric and magnetic field, selected with the
MC-STEPS option followed by the number of collisions over which to
average.
Default settings are:
- time interval: 20 psec
- space interval: 10 micron
- steps to be skipped: 100
Sets the trap distance (in terms of wire radii). If an electron or
ion can be attracted by a wire (this depends on the charge on the
wire but also on the setting of the option CHECK-ALL-WIRES) and if
the particle passes closer by the wire than a distance of
ntrap*radius
then the electron or ion is considered to be caught by the wire.
From the moment a wire is considered caught by a wire, a dedicated
integration algorithm takes over which is better at estimating the
residual drift time than the default algorithm.
[This parameter is preset to a value of the order of 2-5 (depending
on program version). This can be too large if the wires are very
thick but it may as well be too small for very thin wires.]
Depending on their charge, wire can either attract a particle,
repel it or have no effect. Wires can also have a multipole moment
which makes them attractive from one side and not from another.
If the CHECK-ALL-WIRES option is in effect, then all wires, no
matter their charge, are considered able to catch a particle. As
soon as a particle comes closer to any wire than the trap radius
(see: ntrap) an attempt will be made to drift it to the wire.
This is meaningful if you have e.g. dipole (q=0) type wires, but
this is harmful if you particles pass near repelling wires, such
as gating grids. When not needed, this option also wastes a lot
of CPU time.
[This option is on by default.]
When the CHECK-ATTRACTING-WIRES option is in effect, a particle
will not be considered caught when it comes closer than the trap
radius to a wire that is charged such that it can not attract
the particle that is drifting.
This is usually the recommended mode but there are cases, such
as the presence of wires with almost no net charge, but with a
multipole moment, where the alternative is better suited.
[This option is NOT default.]
This option requests drift line calculation to be aborted if
the drift line makes a bend sharper than 90 degrees. Such bends
rarely occur in smooth fields, the most common case is a drift
line that tries to cross a saddle point. The REJECT-KINKS option
will ensure that the drift line doesn't repeatedly go back and
forth across the saddle point.
Since fields obtained with finite element methods occasionally
have areas with very uneven fields, it may be advisable in such
cases to switch the option off.
[The option is on by default.]
Sets the distance from the wire (in multiples of the wire radii)
at which the integration routine for combined longitudinal and
transverse diffusion changes from accumulating the diffusion
covariance matrix to projecting the accumulated probability
distribution onto the target wire.
This parameter is preset to a value of the order of 2-5.
When both transverse and longitudinal diffusion have been entered
in the gas section, the diffusion is calculated by propagating a
cloud along the drift line, adjusting the dimensions at each step
according to the following phenomena:
- longitudinal stretch due to acceleration and decelaration
of the electrons along the central drift line
- transverse compression due to convergence of drift lines
neighbouring to the central drift line
- additional transverse and longitudinal diffusion at each
step, according to the local diffusion coefficients.
The cloud is considered Gaussian far from the wires.
When the drift line approaches the wire, the cloud as a whole
is projected onto the wire. For this phase, various algorithms
are available put at your disposal:
NO-PROJECTION
No special treatment when approaching the wire, hence
the value of ncloud is not relevant.
INTEGRATION
As soon as the cloud enters the 'ncloud' zone,
the following is done:
Complete integration over the cloud of the local
distance to the wire divided by the local drift
velocity.
The longitudinal diffusion over the remaining
distance to the wire is added to the estimate.
CENTRAL-VELOCITY-INTEGRATION
As soon as the cloud enters the 'ncloud' zone, an
integration similar to INTEGRATION is carried out,
but the drift velocity is always taken to be the
drift velocity at the centre of the cloud.
The longitudinal diffusion over the remaining
distance to the wire is added to the estimate.
This is currently the default method.
LONGITUDINAL-DIMENSION
When the cloud center enters the 'ncloud' zone,
the dimension of the cloud over a line through
cloud center and wire center is taken as measure
of diffusion spread.
The longitudinal diffusion over the remaining
distance to the wire is added to the estimate.
The longitudinal dimension is in principle the
dimension that matters, but in the presence of
a strong magnetic field, the cloud rapidly rotates
near the wire. At the same time, the cloud stretches
to the point of becoming almost one-dimensional.
A small rounding error in the cloud alignment, can
make the dimension along the axis pointing to the
wire, very small.
For this reason, this method is not recommended,
unless the cloud trap radius is very large (in which
case the velocity estimates are likely to be
inaccurate).
LARGEST-DIMENSION
This is similar to LONGITUDINAL-DIMENSION but
the cloud size is taken to be the largest cross
section of the cloud.
The longitudinal diffusion over the remaining
distance to the wire is added to the estimate.
For reasons explained under LONGITUDINAL-DIMENSION,
this method must be considered superior, provided
the cloud-trap radius is small.
A step is subdivided if the difference between the first and
second order estimates differ more than a fraction epsilon of
the total first order estimate without subdivisions.
The default is 1.0E-3.
The stack depth is the maximum number of subdivisions allowed
during the integration, in order to achieve the requested
accuracy.
For diffusion coefficients, the stack depth usually hardly matters
since the coefficient does not make big jumps. For the Townsend
coefficient, which suddenly grows from 0 to an appreciable value,
the stack depth is a critical parameter in the accuracy of the
computation. Although CPU time can go up rapidly with stack depth,
it is a good idea to keep a large value: when not needed, no use
of the stack is made.
Default is MXSTCK, usually set to 20, which is also the maximum.
The smallest permitted value is 1 and this setting will usually
already give a reasonable accuracy. The default stack depth is
large and may result in excessively lengthy computations.
Requests isochrones to be drawn as lines, rather than marked.
When this option is selected, you may also wish to inspect the
settings of the other isochrone related options.
[By default, isochrones are drawn as lines.]
Requests marking the points on the isochrones.
If this option is active, no sorting needs to be done. Hence,
the other isochrone options are ignored. Plotting isochrones
is fast with this option switched on.
[By default, isochrones are drawn as lines.]
Depending on the source of the points that serve to draw
the isochrones, they can be in some definite order or not.
By setting SORT-ISOCHRONES, an attempt is made to order
the points in such a manner that the isochrones appear as
reasonably smooth lines. Any attempt to do so is likely
to fail for certain cases. Moreover, sorting can take a
large amount of computing time - the problem is related
to the notorious "traveling salesperson problem (TSP)".
Garfield, for these reasons uses a simple algorithm to
sort the points on a contour: each contour is classified
as being either linear or arcs. Linear contours are
sorted along the main axis, points on arcs by angle with
respect to the centre of gravity. Arcs that appear to be
nearly full loops are drawn as closed contours, otherwise
as an open arc.
The sorting algorithm by itself is fast - the check on
intersects between isochrones and drift lines in contrast
is fairly time consuming.
Sorting is not useful (hence potentially harmful) when
the drift lines come from a track on which the points are
ordered - which is usually the case. The sort is useful
on the other hand for drift lines starting from wires or
other electrodes.
The SORT-ISOCHRONES option is ignored when MARK-ISOCHRONES
is in effect.
[By default: sort done]
Points on an isochrone are only joined if they are less than a
fraction iso_thr away from each other on the screen. Points
that can not be connected are shown by a marker.
The fraction iso_thr can be set to any value between 0 (only
markers) and 1 (isochrones are only interrupted by drift line
crossings).
Selecting NOISOCHRONE-CONNECTION-THRESHOLD is tantamount to
setting iso_thr to 1.
[Initial setting: 0.2]
When an isochrone appears to be more or less circular, its
points are ordered by increasing angle with respect to the
centre of gravity. If the isochrone, on the other hand, seems
to be more or less linear, the points are ordered along the
longest principal axis of the distribution.
Whether the set is circular or linear is decided by computing
the RMS in the two principal axes of the point distribution.
If the ratio of these two RMS's exceeds iso_aspect, then the
isochrone is assumed to be linear, otherwise circular.
[Initial setting: 3]
Isochrones that appears to be circular (rather than linear,
see ISOCHRONE-ASPECT-RATIO-SWITCH for the conditions under
which this happens) are closed if the largest distance
between 2 points doesn't exceed a fraction iso_loop of the
total length of the isochrone.
[Initial setting: 0.2]
Requests ensuring that drift lines do not cross isochrones.
Such crossings can for instance occur if the drift lines
from a track flow left or right of an intermediate object,
which itself also attracts some electrons, to a wire located
behind the object.
This check is fairly time consuming.
[By default: check done]
Keyword index.
Formatted on 10/11/98.